17:44Corey McClain · Demo
This New ChatGPT Voice Update Changes Everything
Corey McClain puts OpenAI's new voice-to-agent update through a live, unscripted stress test, running an entire video-production skill by voice alone.
September 23rdThe argument in one line.
OpenAI's new ChatGPT feature learns your voice from connected email and documents, but a side-by-side test shows a custom-built voice skill calibrated on real writing samples still sounds more like the writer and reads tighter.
Read if. Skip if.
READ IF YOU ARE…
- A creator or solopreneur who publishes regularly and wants AI drafts that don't read like generic AI wrote them.
- Someone deciding between ChatGPT's built-in writing-style personalization and building a custom voice skill from their own writing samples.
- Anyone curious what practical difference a 'voice skill' calibrated on real writing makes versus a general connected-apps feature.
SKIP IF…
- You're looking for a step-by-step build tutorial — this video demos a comparison, not the construction of a voice skill.
- You don't use ChatGPT or have no interest in AI-generated writing quality.
The full version, fast.
OpenAI shipped a ChatGPT feature that connects to your Gmail, Google Drive, and messaging apps and drafts text meant to sound like you. Creator Corey McClain tests it against his own custom-built 'voice skill' by having Gemini write a passage first, then asking each ChatGPT tool to rewrite it in his voice, three times each. Comparing the outputs line by line in Diffchecker, the built-in feature runs noticeably longer and leans on softer, more qualified phrasing, while the calibrated skill stays tighter and closer to how he actually writes. His takeaway: the new feature is a real upgrade over default ChatGPT, but a skill built from your own writing samples, reduced to explicit rules, still wins.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 01:19
01 · The pitch and the setup
Corey claims OpenAI's new ChatGPT writing-style feature ends AI slop, walks through connecting Gmail, Google Docs, and messaging in the settings, and gets ChatGPT to draft a real email in his own inbox on the first try.
01:19 – 02:21
02 · Designing a fair test
Rather than judge the feature on its own first draft, he has Gemini write a passage first, then runs the same passage through both ChatGPT's writing-style feature and his own voice skill, three times each.
02:21 – 03:31
03 · Bringing in Diffchecker
He explains why comparing the outputs inside ChatGPT itself won't do the differences justice, and switches to Diffchecker to place the runs side by side, counting roughly 169 lines from his skill against around 246-247 from the built-in feature.
03:31 – 06:13
04 · Line-by-line comparison
He walks through matching passages from both outputs sentence by sentence, showing where the built-in feature adds extra qualifiers and softer phrasing while his own skill stays tighter and more direct, and lands on the verdict that the calibrated skill sounds more like him and reads better.
06:13 – 08:03
05 · Why the skill wins, and where to get it
He explains the idea behind his voice skill: enough writing samples can be reduced to a mathematically calibrated 'voice print,' unlike the built-in feature's inferred tone. He calls the new OpenAI feature worth trying, then points to the Voice Algorithm Factory for building a personal skill, and signs off.
Lines worth screenshotting.
- OpenAI's new ChatGPT writing-style feature reads connected Gmail, Google Docs, and messaging apps to draft text that sounds less like generic AI.
- The feature has no visible skill-selection step in the chat, so testers can't tell it apart from a custom voice skill without comparing outputs directly.
- A fair test of an AI voice-matching tool means feeding it content written by a different AI, not letting it grade its own first draft.
- A custom voice skill trained on real writing samples produced output noticeably shorter, around 169 lines versus roughly 246 for the same passage rewritten by the built-in feature.
- Line-by-line diffing surfaced the tell: the built-in feature adds qualifiers where the calibrated skill cuts straight to the point.
- The creator's theory: with enough writing samples, a person's style reduces to a measurable 'voice print' that can be enforced by explicit rules instead of inferred by a model.
- OpenAI's writing-style feature is a real upgrade over default ChatGPT output, but connecting apps and inferring tone still isn't the same as encoding explicit stylistic rules.
- The comparison ran three times on both tools to rule out one-off luck before declaring the calibrated skill the more consistent match to the creator's actual voice.
Generic voice-matching still loses to a calibrated skill
AI WRITING VOICE
Connecting your email and docs teaches ChatGPT to sound less generic, but only a skill built from your own writing samples and explicit rules reliably nails your actual voice.
- ChatGPT's new writing-style feature reads your connected Gmail and Google Drive to infer tone, which means it's inferring patterns rather than following explicit rules.
- Testing a voice-matching tool honestly means feeding it content it didn't write, like asking one AI to rewrite another AI's draft, not just grading its own first pass.
- A side-by-side diff tool turns a vague 'does this sound like me' judgment into a line-by-line comparison you can actually point to and defend.
- The rewritten version that runs longer and adds more qualifiers is usually the one still leaning on generic AI habits, not the tighter, more decisive version.
- A large enough sample of your own writing can be reduced to explicit stylistic rules, a kind of measurable voice print, instead of left to a model's best guess.
- Free platform features are a reasonable default, but a purpose-built skill calibrated to your own sentences is worth building once you rely on AI drafts regularly.
Terms worth knowing.
- Writing-style feature
- A ChatGPT capability that connects to a user's email, documents, and messaging apps to draft text that mimics how they normally write.
- Voice skill
- A custom-built ChatGPT skill trained on a person's own writing samples to enforce their specific phrasing, structure, and rules rather than a general inferred tone.
- Diff Checker
- A free web tool that highlights the differences between two blocks of text side by side, line by line.
- Voice print
- The creator's term for reducing someone's writing style down to a set of measurable, repeatable rules derived from a large enough sample of their own writing.
Things they pointed at.
Lines you could clip.
00:00
“There's a brand new feature inside of ChatGPT that's supposed to do away with AI slop completely”
06:49
“I had this idea that if you have enough samples of your own writing, then you can reduce your writing to a voice print like a fingerprint through sheer math.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
There's a brand new feature inside of ChatGPT that's supposed to do away with AI slop completely because we know that ChatGPT has suffered in this area for a very long time. It's supposed to connect to your messaging, your documents, your email, and learn the way you talk, speak, and write, and then write like you. I don't know where to go in the settings to set it up, but I'll put this link in the description that OpenAI left us so that you can go through the setup as well.
My Gmail and my Google Docs are already connected, but let's click on messaging to see what options we have. I don't have Slack and I don't have Teams, so let's back up and then let's choose use writing style. And it doesn't seem to do anything special.
It just says find a recent email or thread and draft a response using my writing style. And after a minute and 13 seconds, it wrote an email. And I think this email is actually inside my inbox.
So we see the chat GPT sits was able to actually write an email. So I know that the system is working. It's reading my emails, my Google drive, and it has enough information to purportedly write in my voice and in my style.
and now we're going to test it and so what we're going to do to test this is instead of giving it some of my writing we're going to let gemini write so it wrote about the same thing short medium and long but now we're going to ask chat gpt to use that new feature to rewrite this gemini content and we're going to see if chat gpt can write it so that it sounds like me So the first thing I did was to ask ChatGPT to rewrite this piece of content using the new writing style features that OpenAI released today.
Do not use any skills or anything else. Use what you've learned about my voice from Gmail and Google Drive. And I use GPT -6 Astra Lite, which is amazing.
Then we got these three pieces of content. And something that you'll notice about each of these pieces of content is that there is no down arrow next to either of these content pieces of content. And the reason why is because ChatGPT did not use the skill.
It used the new feature and what it knows about my writing style from Google Drive and from Gmail. and then we got this one where i asked chat gpt use my corey mclean voice skill and nothing else to rewrite the following passage we did it three times and every time you're going to see where chat gpt worked for a certain amount of time this is where you use the actual skill did the same thing here and also here but trying to compare these inside of chat GPT is not going to do it justice.
And it's probably going to be a little bit boring. So what we're going to do is use this website tonight called diff checker and diff checker is going to allow us to place the copy side by side for each of these runs and look at what changed. So this is the first run where Jim and I gave us the text on the left and it is 169 lines.
If my memory serves me right, 160. 169 lines, maybe 171 if you count down here, but 169 is a number for the beginning of the last lines. And then over here on the right is the copy that the chat GPT feature wrote.
And if we scroll past the bottom, we can see that there are a lot more lines that chat GPT actually wrote trying to sound like me. Then if we come to this tab and we scroll all the way down on the left hand side, we see that the last line begins with 169. And then if we look over here to the right, it keeps going, but it stops at 197 lines.
And that is much shorter than the 246, 247 lines that appear on this page. But let's look at some of the actual things that it rewrote. So let's look right here at number one.
It banishes the blank page. The hardest part of creating is often just starting. AI is the ultimate brainstorming partner that never gets tired.
Whether you need 10 catchy YouTube titles, an outline for a newsletter, or a fresh angle on a saturated topic, you can prompt an AI to generate the raw material. You don't have to use exactly what it spits out. You just need to get the momentum going so you can apply your unique voice.
Sounds good. But let's see what chat GPT wrote. Give yourself something to work with.
Sometimes you know what you want to say, but you haven't figured out how to say it yet. Give AI the idea. Tell it who you're talking to and what you want that person to understand.
Ask for a few directions you could take. You might look at the first response and say, no, that's not it. But this part right here, we can work with that.
Now you have a starting point. You can respond, make changes and build on it. I'll also just talk through an idea and let AI help me organize it.
The thinking is already there. I need help getting it out into a shape somebody else can follow. Now let's look on this one where my voice skill rewrote all of this content.
Give yourself something to work with. Sometimes you know what you want to say. You just don't know where to start.
So give AI the idea. tell it who you're talking to and what you want that person to understand ask for a few titles an outline or another way to explain the point you're going to read through it some of it won't fit that's fine pick the part that gets you thinking and build from there the point is to get started you still decide what you're going to say chat gpt did a pretty good job of trying to sound like me but it didn't quite make it.
So for instance, ChatGPT starts off, give yourself something to work with. And so does my skill. But then ChatGPT says, sometimes you know what you want to say, but you haven't figured out how to say it yet.
My skill says, sometimes you know what you want to say, period. You just don't know where to start, period. So give AI the idea.
And ChatGPT says, give AI the idea. Tell it who you're talking to and what you want that person to understand. Ask for a few directions you could take.
So give AI the idea. Tell it who you're talking to and what you want that person to understand. Ask for a few titles, an outline, or another way to explain the point.
Do you see that? Ask for a few titles, an outline, or another way to explain the point. Ask for a few directions you could take.
it's obvious which one sounds like me the second one my skill and even if you don't think it sounds like me you have to admit it it's definitely better written my voice skill can Completely outperforms anything that I know of. And I've already known this.
And the reason why is very simple. It's not because the chat GPT feature is terrible. It writes a little bit like me, or at the very least, it makes the writing sound a little human.
But if I come back to this page where I rewrote those same three passages with my own voice skill, you will notice that there is this section at the very end of each one of those passages. I had this idea that if you have enough samples of your own writing, then you can reduce your writing to a voice print like a fingerprint through sheer math.
And it doesn't have this small personal ruler like my skill has that makes certain. that evaluates everything that's written so that it sounds like my voice and it's mathematically calibrated. So if you want to try a new feature out, I'll put a link in the description because it's definitely worth it.
It's definitely not slop anymore. I can say that much. But if you really want your voice and your writing to sound like you, then I'm going to put a link in the description where you can download the voice algorithm factory and you can generate your own skill so that all of your writing.
sounds like you absolutely free. If you got value out of the video, make sure you hit the like button, subscribe to the channel and as always take care, have a great day and I'll see you in the next video.
The Hook
The bait, then the rug-pull.
OpenAI just shipped a ChatGPT feature that reads your email and documents to write in your voice. Corey McClain puts it up against a custom voice skill he built himself, and the difference shows up the moment you diff the two outputs line by line.
Named ideas worth stealing.
06:49concept
Voice Print (Personal Ruler)
The idea that with a large enough sample of someone's own writing, their style can be reduced to a set of measurable, repeatable rules, calibrated almost like a fingerprint, rather than approximated by a general model's inferred tone.
Steal forany AI-writing workflow where consistency across many pieces matters more than one good draft
How they asked for the click.
VERBAL ASK
07:47link
“I'm going to put a link in the description where you can download the voice algorithm factory and you can generate your own skill so that all of your writing sounds like you.”
Delivered as the payoff right after the comparison lands, not a hard sell up front; paired with a standard like/subscribe ask in the closing seconds.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
